Sie sind auf Seite 1von 9

Electrical Engineering in Japan, Vol. 174, No.

1, 2011
Translated from Denki Gakkai Ronbunshi, Vol. 128-B, No. 10, October 2008, pp. 1235–1242

A Multibranch Exchange Method for Distribution Loss Minimization

MASAHIKO MURAI,1 YOKO KOSAKA,1 TAKENORI KOBAYASHI,1


and YUMIKO KANESHIGE2
1
Power and Industrial Systems R&D Center, Toshiba Corporation, Japan
2
Power Control Computer Systems Department, Toshiba Corporation, Japan

SUMMARY change method [2], the loss reduction achieved by branch


exchange is estimated without the power flow calculation,
This paper presents a multibranch exchange method which allows fast solution search. However, this is a local
for reconfiguration of distribution systems to reduce their search in the neighborhood formed by branch exchange,
line losses. In this method several switches are closed and and hence only local solutions close to the initial solution
opened simultaneously in each branch exchange operation can be obtained.
to expand the search neighborhood. The switches to be On the other hand, in sequential branch opening, all
closed are selected as the intermediate systems will be switches are closed to configure a meshed network as the
meshed configuration. Sequential branch opening method initial state, after which the power flow is calculated and a
can be applied to open the switches to obtain the radial switch which has minimum current is opened, with repeti-
configuration efficiently. Test examples show the effective- tion of the procedure until a radial configuration is obtained.
ness of the proposed method in the case of double branch This method is known to offer quasi-optimal solutions not
exchange. © 2010 Wiley Periodicals, Inc. Electr Eng Jpn, depending on the initial solution [5]. Feasible solutions are
174(1): 40–48, 2011; Published online in Wiley Online not found until the final radial configuration is obtained,
Library (wileyonlinelibrary.com). DOI 10.1002/eej.21015 and therefore the method may prove impractical when
computing time is limited.
In this paper, we propose a method of loss minimiza-
Key words: distribution system; loss minimiza- tion in a distribution system by multibranch exchange,
tion; combinatorial optimization; heuristics branch ex- which is an extension of the branch exchange method
change; sequential branch opening. involving multiple branches simultaneously in each opera-
tion. Previously, multistage branch exchange was proposed
for distribution system expansion planning by repeated
1. Introduction branch exchange in several stages [6]. The objective func-
tion need not necessarily decrease at every stage of branch
Loss minimization in distribution systems is known exchange, and as a result, the search neighborhood is ex-
to be a combinatorial optimization problem of determining panded to minimize the objective function. However, when
switches on or off so as to minimize losses. However, actual such branch exchange is performed in multiple stages, all
large-scale problems involving thousands or tens of thou- initial and intermediate combinations that do not reduce
sands of switches are difficult to solve by exact methods losses immediately should be checked for final loss reduc-
such as the branch and bound algorithm, and hence approxi- tion. This may complicate the algorithm and result in a
mate heuristic algorithms have been proposed, such as considerable increase of the computing time.
branch exchange [1–3] and sequential branch opening [4, On the other hand, in the multibranch exchange pro-
5]. posed in this study, multiple switches are closed at once to
In the branch exchange method, a radial configura- obtain a meshed configuration; after that, the same number
tion is considered as the initial state. A switch is closed and of switches are opened to obtain a radial configuration
another switch is opened, while maintaining the radial again; thus, the search neighborhood is expanded to mini-
configuration in each operation. The operation is repeated mize the objective function. To obtain a radial configuration
sequentially so as to minimize losses. In the branch ex- from the meshed configuration, usually the switches to be

© 2010 Wiley Periodicals, Inc.


40
opened are selected by the sequential branch opening
(1)
method, which is expected to improve calculation effi-
ciency.
The present paper is organized as follows. First the Here Iin is the section inflow current and Il is the section
problem of distribution loss minimization is formulated in load.
Section 2, and then the multibranch exchange method is Consider the loss minimization problem for the ex-
introduced in Section 3. After that, numerical simulations ample system shown in Fig. 1. In the diagram, F1 to F3 are
using the double-branch exchange are applied to a simple feeders, 1 to 6 are sections, and 1–5 are switches. Open
example system and a real-scale system in Section 4. The and closed switches are represented by " and E, respec-
efficiency of the proposed method is verified by compari- tively. As is obvious from the diagram, the distribution
son with the single-branch exchange. system has a radial configuration, with section 1 supplied
from feeder F1, sections 2 to 4 supplied from feeder 2, and
2. Distribution Loss Minimization Problem sections 5 and 6 supplied from feeder F3.
The power loss is minimized as follows. A switch,
The problem of loss minimization in a distribution such as switch 1, is closed while switch 2 is opened, a new
system consists of choosing switches on or off so as to radial configuration is created, and the loss is calculated.
minimize power losses, while maintaining a radial configu- This procedure is repeated for every possible radial con-
ration. figuration, and the one resulting in the minimum loss is
In practice, one must consider various constraints, selected. There are 8 radial configurations with different
such as voltage limitation and line capacity limitation. As switch states, including that shown in Fig. 1, and hence the
regards voltage, regulation is usually possible by means of best one (with minimum loss) can be found by simple
devices such as transformer taps or SVR. On the other hand, comparison. However, when this algorithm is applied to a
constraints on line capacity limitation are unlikely to be real-scale distribution system, the number of configurations
violated because the load is allocated nearly uniformly (switch combinations) grows enormously, and selection
among feeders in the course of loss minimization. Thus, we becomes impractical. Thus, various methods have been
ignore the above constraints in this study. proposed to find the approximate minimum-loss configu-
We make the following assumptions about the distri- ration within a realistic time frame.
bution system.

• The loads are constant-current loads. 3. Optimization Algorithm


• The load is distributed uniformly in a section.
• The load power factor is 1.
3.1 Algorithm outline
• The current phase change caused by line imped-
ance is negligible. In conventional branch exchange, one of the open
• The switch impedance is negligible. switches in a given radial configuration is closed to make a
new loop, and another is opened so as to minimize the
Under the above assumptions, sectional power loss power loss in the new loop. This procedure is applied to all
can generally be expressed as follows (see Appendix 1): of the open switches, and the lowest loss configuration is
selected. The same procedures are applied again to the
configuration yielding the lowest loss, and so on, in an
attempt to minimize the total power loss in the distribution
system.
On the other hand, in the multibranch exchange
method proposed in this paper, the search neighborhood is
expanded by simultaneous exchange of multiple branches
so that the local optimal solution approaches the global
optimal solution. In particular, the search neighborhood can
be extended by selecting a meshed configuration including
several closed branches. Then sequential branch opening is
applied to this meshed configuration, and quasi-optimal
solutions are obtained rapidly by sequential opening.
The algorithm of multibranch exchange is illustrated
in Fig. 2. The steps of the algorithm are explained below.
Fig. 1. Example distribution system.

41
It should be noted that in Steps 2 to 5, the N switches
to be closed are restricted to switch combinations in which
N loops overlap to form a meshed configuration.
Below we consider the calculation of the switch
currents required for determination of the switches to be
opened in Step 3 so as to obtain a meshed configuration,
and the calculation of the line currents in a meshed configu-
ration required in order to find the power loss at Step 4.

3.2 Calculation of loss minimization in branch


exchange method

First we demonstrate that the branch exchange pro-


cedure is performed by opening switches with minimum
currents in a loop obtained by switch closing.
The amount of loss change caused by branch ex-
change can be expressed as follows [2]:
(2)
Here ∆I is the load transferred between feeders, E1 and E1g
are the voltage drops from respective feeder through the
ends of the switch to be closed, and Rloop is the total
impedance of the loop created by closing switches.
The above also applies to uniformly distributed loads;
a derivation is given in Appendix 2. Since ∆I in Eq. (2) is a
quadratic function, the minimum value of the power loss
achieved by branch exchange occurs when ∆I is the closest
to the following load:
Fig. 2. Flow diagram of multibranch exchange method.
(3)
(Step 1) A radial initial configuration is prepared and
set as a candidate solution, and the power loss of the initial According to Thévenin’s theorem, Eq. (3) is equivalent to the
system is calculated. This power loss is compared to that of variation of the current due to switch closing, and therefore
the modified configurations in Steps 4 and 6. opening the switch with the smallest current in the loop system
(Step 2) N open switches are closed to obtain a is sufficient to transfer the load closest to Eq. (3).
meshed configuration. In the multibranch exchange method, switches are
(Step 3) In the meshed configuration obtained in Step opened so as to sequentially minimize the current in a
2, N switches are opened to obtain a radial configuration, meshed system, and therefore local optimality is not guar-
while sequentially minimizing the current. anteed; however, one can expect quasi-optimal local solu-
(Step 4) The power loss is calculated for the new tions.
radial configuration obtained in Step 3. If the loss is lower
than that of the candidate solution, the solution is updated. 3.3 Calculation of switch current in meshed
(Step 5) A check is made to determine whether N configuration
unprocessed open switches exist. If such switches exist, the
algorithm returns to Step 2; otherwise, it proceeds to Step We next explain the calculation of the line current
6. after the closing of switches, which is required in order to
(Step 6) If the candidate solution has been updated determine the switches whose closure yields the minimum
(with respect to the initial configuration), then the algo- current in a meshed configuration. We consider the ex-
rithm proceeds to Step 7; otherwise, it jumps to Step 8. change of two branches, but the same approach applies to
(Step 7) The candidate solution is considered as a new three or more branches.
initial configuration, and the algorithm returns to Step 2. Figure 3 shows a distribution system in which a loop
(Step 8) The current configuration is output as the is formed by closing one switch. By Thévenin’s theorem,
minimum-loss solution. the amount of current change can be calculated as follows:

42
Fig. 3. Variation of current with first switch closing.

(4)

Here EA and EB are the voltages on, respectively, the feeder


fA and feeder fB sides of the switch; and Rloop = RA + RB,
where RA and RB are the total section impedance from Fig. 4. Variation of current with second switch closing.
switch’s end to feeder fA and feeder fB, respectively.
The current increases and decreases by ∆I on, respec-
tively, the feeder fA and feeder fB sides of the switch. That
(10)
is,
(5) Therefore, the currents in the various parts can be expressed
(6) as follows:
Here IA and IB are the currents on, respectively, the feeder (11)
fA and feeder fB sides of the switch.
The section voltage drop is as follows [see Eq. (A.4) (12)
in Appendix 2]:
(13)
(7) (14)

Here Iin is the section inflow current and Il is the section


3.4 Calculation of currents produced by
load.
opening switches in meshed configuration
Therefore, the voltages EA and EB on both sides of the
switch can be found by applying Eq. (7) to the respective Here we explain the calculation of the line currents
sections from the feeder’s outlet to the end of the switch. when switches are opened in a meshed system, which is
After the switch is closed, the voltage at the next switch to performed in order to find the power loss in a new radial
be closed can be found by applying the same calculation to system.
the updated currents. Figure 5 shows the distribution system obtained by
Figure 4 shows a distribution system in which a loop opening first switch 1. Here the current variation is
is formed by closing switch 2 (second closing). Here the
amount of current change can be calculated as follows:

(8)

Here EB and EC are the voltages on, respectively, the feeder


fAfB and feeder fC sides of switch 2; and RA, RB, and RC is
the total section impedance from branch point to outlet of,
respectively, feeder fA, fB, and fC.
The variation of the current in a loop system formed
by closing one switch is divided in inverse proportion to the
total section impedance from the branch point to the feed-
ers; that is,

(9)
Fig. 5. Variation of current with first switch opening.

43
Table 1. Section loads and impedances of example
system

Fig. 6. Variation of current with second switch opening.

∆I = −ISW, while the current branched to the rest of the loop


is in inverse proportion to the impedance. Therefore,

(15) switch 4, the current passing through the loop can be


(16) calculated as shown in Fig. 8(b). Here, too, the smallest
current is obtained by closing switch 4, and hence the
(17) power loss does not change.
Now let us consider double-branch exchange when
(18)
switch 4 is closed in Fig. 8(a) or switch 1 is closed in Fig.
8(b). When switch 1 is closed in Fig. 8(b), the current that
(19)
flows in the system is shown in Fig. 9(a); therefore, the
system shown in Fig. 9(b) is obtained by opening the
(20) smallest-current switch 2. After opening smallest-current
switch 5 in Fig. 9(b), the new configuration is obtained.
The current variation ∆I for the next switch opening (Fig. The power loss in the system shown in Fig. 7 is
6) can be calculated by using the current with direction 57.067 W, and the system obtained by double-branch ex-
opposite to ISW as follows: change achieves a loss reduction to 56.267 W. This is the
best of the eight possible configurations in terms of power
(21) loss.
Finally, consider the application of the two-stage
(22)
branch exchange method proposed in Ref. 6. In the two-
stage branch exchange, the first-stage branch exchange is
4. Numerical Examples
performed regardless of whether or not the power loss
decreases, which is confirmed at the second-stage branch
In order to verify the multibranch exchange method,
exchange. Therefore, the switch combinations of the first-
we conducted numerical simulations on the example sys-
and second-branch exchange are as shown in Table 2. Thus,
tem in Fig. 1 and on a real-scale system. In particular, we
all eight combinations (including initial system and those
compared single- and double-branch exchange in terms of
basic efficiency.

4.1 Example system in Fig. 1

First we applied conventional branch exchange to the


example system shown in Fig. 1. The section loads and
impedances of the example system are given in Table 1. The
initial current switch and feeder currents are shown in Fig.
7.
In the branch exchange method, a loop system is
formed by closing switch 1 or 4. Then a new radial
configuration is obtained by opening switches so as to
minimize the current passing through the loop. When
switch 1 is closed [Fig. 8(a)], the smallest current is offered
by switch 1. Therefore, this operation does not alter the
system configuration and power loss. For the closing of Fig. 7. Initial currents of example system.

44
Fig. 8. Branch exchange operations.

Fig. 10. Reduction of power loss in successive


iterations.

obtained at the first-stage exchange) must be searched,


which does not appear highly efficient.

4.2 Real-scale system

Fig. 9. Double-branch exchange operations. A loss minimization calculation was performed for
the 44-feeder distribution system described in Table 3. The
system has 1506 switches, including 218 open switches in
the initial state, and 1332 sections. The power loss in the
Table 2. Two-stage branch exchange operation initial state was 938.2371 kW.
Calculation results are given in Table 4, and Fig. 10
shows how power loss is reduced by successive iterations.
For comparison, the results obtained by the conventional
branch exchange method are also given. Here the CPU time
includes only the time required for branch exchange: data
read and other operations are not considered.
As is evident from Table 4 and Fig. 10, the double-
branch exchange method achieves better loss reduction than
the conventional branch exchange method. As regards com-
puting time, the conventional method requires about 0.8 s,
Table 3. Size of real-scale system while the double-branch exchange method requires about
4.4 s, or 5.5 times as long. However, this is a quite practical
time considering the system size.

5. Conclusions

We have developed a method for loss minimization


in a distribution system by using multibranch exchange, and
Table 4. Calculation results have verified the method by applying it to a simple example
system and a real-scale system.
We may expect further reductions of power loss when
three or more branches are exchanged, which would, how-
ever, result in explosive growth of the computing time. In
the future, we plan to investigate the optimal selection of
branch multiplicity and branch combinations for efficient
calculations.

45
In addition, methods should be developed to deal With the section length and impedance denoted by,
with constraints on the voltage and line capacity that are respectively, L and R, the current flowing at a distance x
ignored in this study (including the operation of voltage from the inflow end is Ix = Iin − (x / L)Il. The impedance of
regulators). a small section ∆x at a distance x from inflow end is ∆r =
The method proposed in this study can also be ap- (R/L)∆x. According to assumptions (3) and (4), the reactive
plied to distribution systems involving dispersed gener- part of the power flow can be ignored, and therefore the
ators, which have been gaining popularity. This, too, is a power loss in the small section ∆x can be expressed as
topic for future research. follows:

REFERENCES (A.1)

1. Ross DW, Patton J, Cohen AI, Carson M. New Therefore, the power loss in this section is
method for evaluating distribution automation and
control (DAC) systems benefits. IEEE Trans Power (A.2)
Apparatus Syst 1981;PAS-100:2978–2986.
2. Civanlar S, Grainger JJ, Yin H, Lee SSH. Distribution
2. Derivation of Power Loss Reduction Due to
feeder reconfiguration for loss reduction. IEEE Trans
Branch Exchange
Power Delivery 1988;3:1217–1223.
3. Baran ME, Wu FF. Network reconfiguration in dis-
tribution systems for loss reduction and load balanc- 2.1 Derivation of power loss increase due to
ing. IEEE Trans Power Delivery 1989;4:1401–1407. load addition
4. Merlin A, Back H. Search for a minimal-loss operat-
ing spanning tree configuration for an urban power Let us find the increase in the power loss when a load
distribution system. Proc 5th PSCC 1975;1:1–18. ∆I is added to the end of the transmission line as shown in
5. Shirmohammadi D, Hong HW. Reconfiguration of Fig. A.2.
electric distribution networks for resistive line losses With the load and impedance of section i denoted by,
reduction. IEEE Trans Power Delivery 1989;4:1492– respectively, Ii and Ri, the increase of the power loss in this
1498. section can be found by using Eq. (A.2) as follows:
6. Nara K, Satoh T, Kuwabara H, Aoki K, Kitagawa M,
Ishihara T. Distribution systems expansion planning
by multi-stage branch exchange. IEEE Trans Power
Syst 1992;7:208–214.

APPENDIX

1. Derivation of Power Loss for Uniformly


Distributed Loss (A.3)
Consider a section with uniformly distributed loss
current Il and inflow current Iin as shown in Fig. A.1. Here ∆Ei is the voltage drop in the section. Here we express
the voltage drop in the section in Fig. A.1 as follows:

(A.4)

Fig. A.1. Uniformly distributed load. Fig. A.2. Distribution line with added load.

46
Therefore, the increase of the power loss in the entire Now the following is obtained by substituting Eq. (A.9) into
distribution system is Eq. (A.7):

(A.5)
2.2 Derivation of Eq. (2)

The load of sections 1 to l is transferred from feeder


F1 to feeder F2 as shown in Fig. A.3.
If branching is present, we may assume that the
(A.10)
branch point is a section with zero impedance and the load
of the branch section because the power loss does not
In addition, ∆PL can be expressed by means of Eq. (A.2) as
change in the branch sections. The variation of the power follows:
loss ∆PLoss can be expressed as follows:

(A.6)

Here ∆PF1, ∆PL, and ∆PF2 are the variations at, respectively,
sections l + 1 to n of feeder F1, sections 1 to l of feeder F1,
and feeder F2.
Using Eq. (A.5), ∆PF1 and ∆PF2 can be expressed as
follows:

(A.7)

(A.8) (A.11)

__
Here ∆Ei is the voltage drop of section i after the load Thus, from Eqs. (A.6), (A.8), (A.10), and (A.11),
transfer; it can be found from the voltage drop before the
load transfer as follows:

(A.12)

Here E1 and E1g are the voltage drops from feeders F1 and
F2 to the ends of the switch to be closed, respectively.
(A.9)
The part of the third term in curly brackets in Eq.
(A.12) can be transformed as follows:

(A.13)

Therefore, the following can be derived from Eqs. (A.12)


and (A.13):

(A.14)
Fig. A.3. Load transfer.

47
AUTHORS (from left to right)

Masahiko Murai (member) completed the M.E. program at Kyoto University in 1991 and joined Toshiba Corporation.
He is now affiliated with the Power and Industrial Systems R&D Center. His research interests are planning, operation and
control of power and energy systems. He is a member of ISCIE, SICE, ORSJ, and IEEE.

Yoko Kosaka (member) received a bachelor’s degree from Nihon University in 1990 and joined Toshiba Corporation. She
is now affiliated with the Power and Industrial Systems R&D Center. Her research interests are operation, supervision, control,
and analysis of power systems.

Takenori Kobayashi (senior member) completed the doctoral program at the University of Tokyo in 1995. He joined
Toshiba Corporation in 1994, and is now affiliated with the Power and Industrial Systems R&D Center. His research interests
are analysis, operation, and control of power and energy systems. He holds a D.Eng. degree, and is a member of IEEE.

Yumiko Kaneshiga (member) received a bachelor’s degree from Ochanomizu University in 1986 and joined Toshiba
Corporation. She is now affiliated with the Power Control Computer Systems Department, and is engaged in the development
of power systems control facilities.

48

Das könnte Ihnen auch gefallen